1. No category

A Benchmark to Evaluate Mobile Video
Upload to Cloud Infrastructures
authors
Afsin Akdogan, HienList
To,of
Seon
Ho Kim, Cyrus Shahabi
Integrated Media Systems Center
University of Southern California
Introduction
How to evaluate the performance of mobile video
applications on these cloud infrastructures and select
an appropriate set of resources for a given application?
Performance Evaluation
Overall Cost Analysis
Other than general purpose instance, the performance
difference between the optimized servers (c, r, i) is not significant
even though the prices vary widely.
Given:
1.  A massive amount of mobile videos’ metadata.
2.  Cloud providers (e.g., Amazon, Google, Microsoft) allow
users to lease computing resources with varying disk,
network and CPU capacities.
Benchmark Design
Upload videos with metadata to cloud has three stages:
1) network to transfer videos from mobile clients to the
cloud servers
2) database to insert metadata about the uploaded videos
3) video transcoding to change the resolution of
uploaded videos to use less storage and bandwidth
Fig. 2a) Comparison of system components
on smallest servers of 4 server types
Fig. 2b) Number of frames that can be
processed for each dollar spent (log-scale).
Transcoding Analysis
Multi-threading: While multi-thread technique suffers from low
parallelism, parallel-single thread can utilize available CPUs better.
We define a single (cross-resource) metric to evaluate the
uploading workflow of video applications on cloud.
Fig. 3) Scale up performance of transcoding (230MB video)
Tab. 1) Categorization of the server types with the prices (dollars/hour)
of the smallest and the largest servers at each group.
Reducing video quality: Throughput increases significantly as the
resolution decreases. However, the percentage improvement
diminishes when the output video resolution becomes too small.
Tab. 2) Transcoding throughput for mp4 and avi types with various output
resolutions. The input video is in .m4v format with 960x540 resolution.
Database Analysis
Tab. 2) Hardware specifications of the smallest and largest servers of 4 types on EC2
MediaQ Architecture
MediaQ: Mobile Multimedia Management System
Fig. 4 ) Insertion throughput on the smallest and largest servers in 4 instance families
Conclusion and Future Work
Fig. 1 ) Overall architecture of MediaQ system
ü  This study identifies the main cost components of a multimedia system
ü  Transcoding is the major bottleneck, which can be more efficient by trading
off the quality the video
ü  In the future, we aim to partition the data across multiple servers and to
provide high scalability
Research Sponsor Logos Here
IMSC Retreat 2015